Statistics
| Branch: | Revision:

ffmpeg / libavcodec / dsputil.h @ 58c2182d

History | View | Annotate | Download (20 KB)

1 ff4ec49e Fabrice Bellard
/*
2
 * DSP utils
3
 * Copyright (c) 2000, 2001, 2002 Fabrice Bellard.
4 8f2ab833 Michael Niedermayer
 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
5 ff4ec49e Fabrice Bellard
 *
6
 * This library is free software; you can redistribute it and/or
7
 * modify it under the terms of the GNU Lesser General Public
8
 * License as published by the Free Software Foundation; either
9
 * version 2 of the License, or (at your option) any later version.
10
 *
11
 * This library is distributed in the hope that it will be useful,
12
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14
 * Lesser General Public License for more details.
15
 *
16
 * You should have received a copy of the GNU Lesser General Public
17
 * License along with this library; if not, write to the Free Software
18
 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
19
 */
20 24641185 Michael Niedermayer
21
/**
22
 * @file dsputil.h
23 983e3246 Michael Niedermayer
 * DSP utils.
24 5755c27f Michael Niedermayer
 * note, many functions in here may use MMX which trashes the FPU state, it is
25
 * absolutely necessary to call emms_c() between dsp & float/double code
26 24641185 Michael Niedermayer
 */
27
28 de6d9b64 Fabrice Bellard
#ifndef DSPUTIL_H
29
#define DSPUTIL_H
30
31
#include "common.h"
32 43f1708f Juanjo
#include "avcodec.h"
33 de6d9b64 Fabrice Bellard
34 24641185 Michael Niedermayer
35 44eb4951 Michael Niedermayer
//#define DEBUG
36 de6d9b64 Fabrice Bellard
/* dct code */
37
typedef short DCTELEM;
38
39 03c94ede Fabrice Bellard
void fdct_ifast (DCTELEM *data);
40 48b1f800 Roman Shaposhnik
void fdct_ifast248 (DCTELEM *data);
41 28db7fce Michael Niedermayer
void ff_jpeg_fdct_islow (DCTELEM *data);
42 10acc479 Roman Shaposhnik
void ff_fdct248_islow (DCTELEM *data);
43 de6d9b64 Fabrice Bellard
44
void j_rev_dct (DCTELEM *data);
45
46 3f09f52a Arpi
void ff_fdct_mmx(DCTELEM *block);
47 cf3bf5bb Michael Niedermayer
void ff_fdct_mmx2(DCTELEM *block);
48 8fd19ab2 Michael Niedermayer
void ff_fdct_sse2(DCTELEM *block);
49 de6d9b64 Fabrice Bellard
50 e0eac44e Fabrice Bellard
/* encoding scans */
51 0c1a9eda Zdenek Kabelac
extern const uint8_t ff_alternate_horizontal_scan[64];
52
extern const uint8_t ff_alternate_vertical_scan[64];
53
extern const uint8_t ff_zigzag_direct[64];
54 10acc479 Roman Shaposhnik
extern const uint8_t ff_zigzag248_direct[64];
55 5a240838 Michael Niedermayer
56 de6d9b64 Fabrice Bellard
/* pixel operations */
57
#define MAX_NEG_CROP 384
58
59
/* temporary */
60 0c1a9eda Zdenek Kabelac
extern uint32_t squareTbl[512];
61
extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP];
62 de6d9b64 Fabrice Bellard
63 44cb64ee Mike Melanson
/* VP3 DSP functions */
64
void vp3_dsp_init_c(void);
65
void vp3_idct_put_c(int16_t *input_data, int16_t *dequant_matrix,
66
    int coeff_count, uint8_t *dest, int stride);
67
void vp3_idct_add_c(int16_t *input_data, int16_t *dequant_matrix,
68
    int coeff_count, uint8_t *dest, int stride);
69
70
void vp3_dsp_init_mmx(void);
71
void vp3_idct_put_mmx(int16_t *input_data, int16_t *dequant_matrix,
72
    int coeff_count, uint8_t *dest, int stride);
73
void vp3_idct_add_mmx(int16_t *input_data, int16_t *dequant_matrix,
74
    int coeff_count, uint8_t *dest, int stride);
75
76 de6d9b64 Fabrice Bellard
77 b7c27ee6 Michael Niedermayer
/* minimum alignment rules ;)
78 eb4b3dd3 Zdenek Kabelac
if u notice errors in the align stuff, need more alignment for some asm code for some cpu
79 b7c27ee6 Michael Niedermayer
or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
80

81
!warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
82
i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
83 de6d9b64 Fabrice Bellard

84 b7c27ee6 Michael Niedermayer
!future video codecs might need functions with less strict alignment
85
*/
86
87 eb4b3dd3 Zdenek Kabelac
/*
88 0c1a9eda Zdenek Kabelac
void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
89
void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
90
void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
91
void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
92 649c00c9 Michael Niedermayer
void clear_blocks_c(DCTELEM *blocks);
93 eb4b3dd3 Zdenek Kabelac
*/
94 de6d9b64 Fabrice Bellard
95
/* add and put pixel (decoding) */
96 b7c27ee6 Michael Niedermayer
// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
97 7d67aa9b Michael Niedermayer
//h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
98 0c1a9eda Zdenek Kabelac
typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
99 669ac79c Michael Niedermayer
typedef void (*tpel_mc_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int w, int h);
100 0c1a9eda Zdenek Kabelac
typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
101 0da71265 Michael Niedermayer
typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
102 b3184779 Michael Niedermayer
103 db794953 Michael Niedermayer
#define DEF_OLD_QPEL(name)\
104 0c1a9eda Zdenek Kabelac
void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
105
void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
106
void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
107 db794953 Michael Niedermayer
108
DEF_OLD_QPEL(qpel16_mc11_old_c)
109
DEF_OLD_QPEL(qpel16_mc31_old_c)
110
DEF_OLD_QPEL(qpel16_mc12_old_c)
111
DEF_OLD_QPEL(qpel16_mc32_old_c)
112
DEF_OLD_QPEL(qpel16_mc13_old_c)
113
DEF_OLD_QPEL(qpel16_mc33_old_c)
114
DEF_OLD_QPEL(qpel8_mc11_old_c)
115
DEF_OLD_QPEL(qpel8_mc31_old_c)
116
DEF_OLD_QPEL(qpel8_mc12_old_c)
117
DEF_OLD_QPEL(qpel8_mc32_old_c)
118
DEF_OLD_QPEL(qpel8_mc13_old_c)
119
DEF_OLD_QPEL(qpel8_mc33_old_c)
120 b3184779 Michael Niedermayer
121
#define CALL_2X_PIXELS(a, b, n)\
122
static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
123
    b(block  , pixels  , line_size, h);\
124
    b(block+n, pixels+n, line_size, h);\
125
}
126 44eb4951 Michael Niedermayer
127 de6d9b64 Fabrice Bellard
/* motion estimation */
128 7d67aa9b Michael Niedermayer
// h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
129
// allthough currently h<4 is not used as functions with width <8 are not used and neither implemented
130 bb198e19 Michael Niedermayer
typedef int (*me_cmp_func)(void /*MpegEncContext*/ *s, uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size, int h)/* __attribute__ ((const))*/;
131 1457ab52 Michael Niedermayer
132 0da71265 Michael Niedermayer
133 24641185 Michael Niedermayer
/**
134
 * DSPContext.
135
 */
136 eb4b3dd3 Zdenek Kabelac
typedef struct DSPContext {
137
    /* pixel ops : interface with DCT */
138 0c1a9eda Zdenek Kabelac
    void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
139
    void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
140
    void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
141
    void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
142 d518aebd Michael Niedermayer
    /**
143
     * translational global motion compensation.
144
     */
145 0c1a9eda Zdenek Kabelac
    void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
146 d518aebd Michael Niedermayer
    /**
147
     * global motion compensation.
148
     */
149 0c1a9eda Zdenek Kabelac
    void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
150 eb4b3dd3 Zdenek Kabelac
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
151
    void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
152 0c1a9eda Zdenek Kabelac
    int (*pix_sum)(uint8_t * pix, int line_size);
153
    int (*pix_norm1)(uint8_t * pix, int line_size);
154 bb198e19 Michael Niedermayer
// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
155
    
156 622348f9 Michael Niedermayer
    me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
157
    me_cmp_func sse[5];
158
    me_cmp_func hadamard8_diff[5];
159
    me_cmp_func dct_sad[5];
160
    me_cmp_func quant_psnr[5];
161
    me_cmp_func bit[5];
162
    me_cmp_func rd[5];
163
    me_cmp_func vsad[5];
164
    me_cmp_func vsse[5];
165 1457ab52 Michael Niedermayer
166 bb198e19 Michael Niedermayer
    me_cmp_func me_pre_cmp[5];
167
    me_cmp_func me_cmp[5];
168
    me_cmp_func me_sub_cmp[5];
169
    me_cmp_func mb_cmp[5];
170 622348f9 Michael Niedermayer
    me_cmp_func ildct_cmp[5]; //only width 16 used
171 eb4b3dd3 Zdenek Kabelac
172 d518aebd Michael Niedermayer
    /**
173
     * Halfpel motion compensation with rounding (a+b+1)>>1.
174 669ac79c Michael Niedermayer
     * this is an array[4][4] of motion compensation funcions for 4 
175 e5771f4f Michael Niedermayer
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
176 5755c27f Michael Niedermayer
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
177 d518aebd Michael Niedermayer
     * @param block destination where the result is stored
178
     * @param pixels source
179
     * @param line_size number of bytes in a horizontal line of block
180
     * @param h height
181
     */
182 669ac79c Michael Niedermayer
    op_pixels_func put_pixels_tab[4][4];
183 d518aebd Michael Niedermayer
184
    /**
185
     * Halfpel motion compensation with rounding (a+b+1)>>1.
186 4e8eed2f Mike Melanson
     * This is an array[4][4] of motion compensation functions for 4 
187 e5771f4f Michael Niedermayer
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
188 5755c27f Michael Niedermayer
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
189 d518aebd Michael Niedermayer
     * @param block destination into which the result is averaged (a+b+1)>>1
190
     * @param pixels source
191
     * @param line_size number of bytes in a horizontal line of block
192
     * @param h height
193
     */
194 da3b9756 Mike Melanson
    op_pixels_func avg_pixels_tab[4][4];
195 d518aebd Michael Niedermayer
196
    /**
197
     * Halfpel motion compensation with no rounding (a+b)>>1.
198 eb14c713 Michael Niedermayer
     * this is an array[2][4] of motion compensation funcions for 2 
199
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
200 5755c27f Michael Niedermayer
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
201 d518aebd Michael Niedermayer
     * @param block destination where the result is stored
202
     * @param pixels source
203
     * @param line_size number of bytes in a horizontal line of block
204
     * @param h height
205
     */
206 eb4b3dd3 Zdenek Kabelac
    op_pixels_func put_no_rnd_pixels_tab[2][4];
207 d518aebd Michael Niedermayer
208
    /**
209
     * Halfpel motion compensation with no rounding (a+b)>>1.
210 eb14c713 Michael Niedermayer
     * this is an array[2][4] of motion compensation funcions for 2 
211
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
212 5755c27f Michael Niedermayer
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
213 d518aebd Michael Niedermayer
     * @param block destination into which the result is averaged (a+b)>>1
214
     * @param pixels source
215
     * @param line_size number of bytes in a horizontal line of block
216
     * @param h height
217
     */
218 eb4b3dd3 Zdenek Kabelac
    op_pixels_func avg_no_rnd_pixels_tab[2][4];
219 669ac79c Michael Niedermayer
    
220 c0a0170c Michael Niedermayer
    void (*put_no_rnd_pixels_l2[2])(uint8_t *block/*align width (8 or 16)*/, const uint8_t *a/*align 1*/, const uint8_t *b/*align 1*/, int line_size, int h);
221
    
222 669ac79c Michael Niedermayer
    /**
223
     * Thirdpel motion compensation with rounding (a+b+1)>>1.
224
     * this is an array[12] of motion compensation funcions for the 9 thirdpel positions<br>
225
     * *pixels_tab[ xthirdpel + 4*ythirdpel ]
226
     * @param block destination where the result is stored
227
     * @param pixels source
228
     * @param line_size number of bytes in a horizontal line of block
229
     * @param h height
230
     */
231
    tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
232 da3b9756 Mike Melanson
    tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
233
234 eb4b3dd3 Zdenek Kabelac
    qpel_mc_func put_qpel_pixels_tab[2][16];
235
    qpel_mc_func avg_qpel_pixels_tab[2][16];
236
    qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
237
    qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
238 1457ab52 Michael Niedermayer
    qpel_mc_func put_mspel_pixels_tab[8];
239 0da71265 Michael Niedermayer
    
240
    /**
241
     * h264 Chram MC
242
     */
243
    h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
244
    h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
245 eb4b3dd3 Zdenek Kabelac
246 0da71265 Michael Niedermayer
    qpel_mc_func put_h264_qpel_pixels_tab[3][16];
247
    qpel_mc_func avg_h264_qpel_pixels_tab[3][16];
248
    
249 bb198e19 Michael Niedermayer
    me_cmp_func pix_abs[2][4];
250 11f18faf Michael Niedermayer
    
251
    /* huffyuv specific */
252
    void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
253 1457ab52 Michael Niedermayer
    void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
254 84705403 Michael Niedermayer
    /**
255
     * subtract huffyuv's variant of median prediction
256
     * note, this might read from src1[-1], src2[-1]
257
     */
258
    void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
259 3d2e8cce Michael Niedermayer
    void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w);
260 b0368839 Michael Niedermayer
    
261 332f9ac4 Michael Niedermayer
    void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
262
    void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
263
264 b0368839 Michael Niedermayer
    /* (I)DCT */
265
    void (*fdct)(DCTELEM *block/* align 16*/);
266 10acc479 Roman Shaposhnik
    void (*fdct248)(DCTELEM *block/* align 16*/);
267 4fb518c3 Michael Niedermayer
    
268
    /* IDCT really*/
269
    void (*idct)(DCTELEM *block/* align 16*/);
270 24641185 Michael Niedermayer
    
271
    /**
272 77c92c2d Michael Niedermayer
     * block -> idct -> clip to unsigned 8 bit -> dest.
273 24641185 Michael Niedermayer
     * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
274 a1adf436 Michael Niedermayer
     * @param line_size size in bytes of a horizotal line of dest
275 24641185 Michael Niedermayer
     */
276 b0368839 Michael Niedermayer
    void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
277 24641185 Michael Niedermayer
    
278
    /**
279
     * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
280 a1adf436 Michael Niedermayer
     * @param line_size size in bytes of a horizotal line of dest
281 24641185 Michael Niedermayer
     */
282 b0368839 Michael Niedermayer
    void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
283 24641185 Michael Niedermayer
    
284
    /**
285 77c92c2d Michael Niedermayer
     * idct input permutation.
286 05493021 Michael Niedermayer
     * several optimized IDCTs need a permutated input (relative to the normal order of the reference
287
     * IDCT)
288
     * this permutation must be performed before the idct_put/add, note, normally this can be merged
289
     * with the zigzag/alternate scan<br>
290 24641185 Michael Niedermayer
     * an example to avoid confusion:
291
     * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
292
     * - (x -> referece dct -> reference idct -> x)
293
     * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
294
     * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
295
     */
296 b0368839 Michael Niedermayer
    uint8_t idct_permutation[64];
297
    int idct_permutation_type;
298
#define FF_NO_IDCT_PERM 1
299
#define FF_LIBMPEG2_IDCT_PERM 2
300
#define FF_SIMPLE_IDCT_PERM 3
301
#define FF_TRANSPOSE_IDCT_PERM 4
302
303 364a1797 Michael Niedermayer
    int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
304
    void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
305
#define BASIS_SHIFT 16
306
#define RECON_SHIFT 6
307
308 44cb64ee Mike Melanson
    /**
309
     * This function handles any initialization for the VP3 DSP functions.
310
     */
311
    void (*vp3_dsp_init)(void);
312
313
    /** 
314
     * This function is responsible for taking a block of zigzag'd,
315
     * quantized DCT coefficients, reconstructing the original block of
316
     * samples, and placing it into the output.
317
     * @param input_data 64 zigzag'd, quantized DCT coefficients
318
     * @param dequant_matrix 64 zigzag'd quantizer coefficients
319
     * @param coeff_count index of the last coefficient
320
     * @param dest the final output location where the transformed samples
321
     * are to be placed
322
     * @param stride the width in 8-bit samples of a line on this plane
323
     */
324
    void (*vp3_idct_put)(int16_t *input_data, int16_t *dequant_matrix,
325
        int coeff_count, uint8_t *dest, int stride);
326
327
    /** 
328
     * This function is responsible for taking a block of zigzag'd,
329
     * quantized DCT coefficients, reconstructing the original block of
330
     * samples, and adding the transformed samples to an existing block of
331
     * samples in the output.
332
     * @param input_data 64 zigzag'd, quantized DCT coefficients
333
     * @param dequant_matrix 64 zigzag'd quantizer coefficients
334
     * @param coeff_count index of the last coefficient
335
     * @param dest the final output location where the transformed samples
336
     * are to be placed
337
     * @param stride the width in 8-bit samples of a line on this plane
338
     */
339
    void (*vp3_idct_add)(int16_t *input_data, int16_t *dequant_matrix,
340
        int coeff_count, uint8_t *dest, int stride);
341
342 eb4b3dd3 Zdenek Kabelac
} DSPContext;
343
344 59cf08ce Fabrice Bellard
void dsputil_static_init(void);
345 b0368839 Michael Niedermayer
void dsputil_init(DSPContext* p, AVCodecContext *avctx);
346 de6d9b64 Fabrice Bellard
347 7801d21d Michael Niedermayer
/**
348
 * permute block according to permuatation.
349
 * @param last last non zero element in scantable order
350
 */
351 0c1a9eda Zdenek Kabelac
void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
352 e0eac44e Fabrice Bellard
353 622348f9 Michael Niedermayer
void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
354
355 d8085ea7 Michael Niedermayer
#define        BYTE_VEC32(c)        ((c)*0x01010101UL)
356
357
static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
358
{
359
    return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
360
}
361
362
static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
363
{
364
    return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
365
}
366
367 24641185 Michael Niedermayer
/**
368 77c92c2d Michael Niedermayer
 * Empty mmx state.
369 24641185 Michael Niedermayer
 * this must be called between any dsp function and float/double code.
370
 * for example sin(); dsp->idct_put(); emms_c(); cos()
371
 */
372 eb4b3dd3 Zdenek Kabelac
#define emms_c()
373
374 e629ab68 Romain Dolbeau
/* should be defined by architectures supporting
375
   one or more MultiMedia extension */
376
int mm_support(void);
377
378 3d03c0a2 Fabrice Bellard
#if defined(HAVE_MMX)
379 de6d9b64 Fabrice Bellard
380 18f77016 Zdenek Kabelac
#undef emms_c
381 eb4b3dd3 Zdenek Kabelac
382 de6d9b64 Fabrice Bellard
#define MM_MMX    0x0001 /* standard MMX */
383
#define MM_3DNOW  0x0004 /* AMD 3DNOW */
384
#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
385
#define MM_SSE    0x0008 /* SSE functions */
386
#define MM_SSE2   0x0010 /* PIV SSE2 functions */
387
388
extern int mm_flags;
389
390 0c1a9eda Zdenek Kabelac
void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
391
void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
392 de6d9b64 Fabrice Bellard
393
static inline void emms(void)
394
{
395 fb16b7e7 Fabrice Bellard
    __asm __volatile ("emms;":::"memory");
396
}
397
398 1457ab52 Michael Niedermayer
399 fb16b7e7 Fabrice Bellard
#define emms_c() \
400
{\
401
    if (mm_flags & MM_MMX)\
402
        emms();\
403 de6d9b64 Fabrice Bellard
}
404
405
#define __align8 __attribute__ ((aligned (8)))
406
407 b0368839 Michael Niedermayer
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
408
void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
409 2720569a Zdenek Kabelac
410 3d03c0a2 Fabrice Bellard
#elif defined(ARCH_ARMV4L)
411
412
/* This is to use 4 bytes read to the IDCT pointers for some 'zero'
413
   line ptimizations */
414
#define __align8 __attribute__ ((aligned (4)))
415
416 b0368839 Michael Niedermayer
void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
417 3d03c0a2 Fabrice Bellard
418 c34270f5 Fabrice Bellard
#elif defined(HAVE_MLIB)
419
420
/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
421
#define __align8 __attribute__ ((aligned (8)))
422
423 b0368839 Michael Niedermayer
void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
424 c34270f5 Fabrice Bellard
425 1e98dffb Nick Kurshev
#elif defined(ARCH_ALPHA)
426
427
#define __align8 __attribute__ ((aligned (8)))
428
429 b0368839 Michael Niedermayer
void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
430 1e98dffb Nick Kurshev
431 59925ef2 Brian Foley
#elif defined(ARCH_POWERPC)
432
433 404d2241 Brian Foley
#define MM_ALTIVEC    0x0001 /* standard AltiVec */
434
435
extern int mm_flags;
436
437 3b991c54 Romain Dolbeau
#if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
438 9a197a24 Michael Niedermayer
#define pixel altivec_pixel
439 3b991c54 Romain Dolbeau
#include <altivec.h>
440 9a197a24 Michael Niedermayer
#undef pixel
441 3b991c54 Romain Dolbeau
#endif
442
443 59925ef2 Brian Foley
#define __align8 __attribute__ ((aligned (16)))
444
445 b0368839 Michael Niedermayer
void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
446 59925ef2 Brian Foley
447 d46aba26 Leon van Stuivenberg
#elif defined(HAVE_MMI)
448
449
#define __align8 __attribute__ ((aligned (16)))
450
451 b0368839 Michael Niedermayer
void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
452 d46aba26 Leon van Stuivenberg
453 0c6bd2ea BERO
#elif defined(ARCH_SH4)
454
455
#define __align8 __attribute__ ((aligned (8)))
456
457
void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
458
459 de6d9b64 Fabrice Bellard
#else
460
461
#define __align8
462
463
#endif
464
465 6d4985bb Fabrice Bellard
#ifdef __GNUC__
466
467
struct unaligned_64 { uint64_t l; } __attribute__((packed));
468
struct unaligned_32 { uint32_t l; } __attribute__((packed));
469 669ac79c Michael Niedermayer
struct unaligned_16 { uint16_t l; } __attribute__((packed));
470 6d4985bb Fabrice Bellard
471 669ac79c Michael Niedermayer
#define LD16(a) (((const struct unaligned_16 *) (a))->l)
472 6d4985bb Fabrice Bellard
#define LD32(a) (((const struct unaligned_32 *) (a))->l)
473
#define LD64(a) (((const struct unaligned_64 *) (a))->l)
474
475
#define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
476
477
#else /* __GNUC__ */
478
479 669ac79c Michael Niedermayer
#define LD16(a) (*((uint16_t*)(a)))
480 6d4985bb Fabrice Bellard
#define LD32(a) (*((uint32_t*)(a)))
481
#define LD64(a) (*((uint64_t*)(a)))
482
483
#define ST32(a, b) *((uint32_t*)(a)) = (b)
484
485
#endif /* !__GNUC__ */
486
487 43f1708f Juanjo
/* PSNR */
488 0c1a9eda Zdenek Kabelac
void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
489 43f1708f Juanjo
              int orig_linesize[3], int coded_linesize,
490
              AVCodecContext *avctx);
491 bb6f5690 Fabrice Bellard
492
/* FFT computation */
493
494
/* NOTE: soon integer code will be added, so you must use the
495
   FFTSample type */
496
typedef float FFTSample;
497
498
typedef struct FFTComplex {
499
    FFTSample re, im;
500
} FFTComplex;
501
502
typedef struct FFTContext {
503
    int nbits;
504
    int inverse;
505
    uint16_t *revtab;
506
    FFTComplex *exptab;
507
    FFTComplex *exptab1; /* only used by SSE code */
508
    void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
509
} FFTContext;
510
511 68951ecf Gildas Bazin
int ff_fft_init(FFTContext *s, int nbits, int inverse);
512
void ff_fft_permute(FFTContext *s, FFTComplex *z);
513
void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
514
void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
515
void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
516 8d268a7d Fabrice Bellard
517 68951ecf Gildas Bazin
static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
518 bb6f5690 Fabrice Bellard
{
519
    s->fft_calc(s, z);
520
}
521 68951ecf Gildas Bazin
void ff_fft_end(FFTContext *s);
522 bb6f5690 Fabrice Bellard
523
/* MDCT computation */
524
525
typedef struct MDCTContext {
526
    int n;  /* size of MDCT (i.e. number of input data * 2) */
527
    int nbits; /* n = 2^nbits */
528
    /* pre/post rotation tables */
529
    FFTSample *tcos;
530
    FFTSample *tsin;
531
    FFTContext fft;
532
} MDCTContext;
533
534 82696bee Fabrice Bellard
int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
535 eb4b3dd3 Zdenek Kabelac
void ff_imdct_calc(MDCTContext *s, FFTSample *output,
536 bb6f5690 Fabrice Bellard
                const FFTSample *input, FFTSample *tmp);
537 eb4b3dd3 Zdenek Kabelac
void ff_mdct_calc(MDCTContext *s, FFTSample *out,
538 bb6f5690 Fabrice Bellard
               const FFTSample *input, FFTSample *tmp);
539 82696bee Fabrice Bellard
void ff_mdct_end(MDCTContext *s);
540 bb6f5690 Fabrice Bellard
541 bb198e19 Michael Niedermayer
#define WARPER8_16(name8, name16)\
542
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
543
    return name8(s, dst           , src           , stride, h)\
544
          +name8(s, dst+8         , src+8         , stride, h);\
545
}
546
547
#define WARPER8_16_SQ(name8, name16)\
548
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
549
    int score=0;\
550
    score +=name8(s, dst           , src           , stride, 8);\
551
    score +=name8(s, dst+8         , src+8         , stride, 8);\
552
    if(h==16){\
553
        dst += 8*stride;\
554
        src += 8*stride;\
555
        score +=name8(s, dst           , src           , stride, 8);\
556
        score +=name8(s, dst+8         , src+8         , stride, 8);\
557
    }\
558
    return score;\
559 1457ab52 Michael Niedermayer
}
560
561 95e2ce4a Michael Niedermayer
#ifndef HAVE_LRINTF
562 9d85cbd9 Fabrice Bellard
/* XXX: add ISOC specific test to avoid specific BSD testing. */
563
/* better than nothing implementation. */
564 6234d753 Alex Beregszaszi
/* btw, rintf() is existing on fbsd too -- alex */
565 9d85cbd9 Fabrice Bellard
static inline long int lrintf(float x)
566
{
567 ea937d01 Fabrice Bellard
#ifdef CONFIG_WIN32
568
    /* XXX: incorrect, but make it compile */
569
    return (int)(x);
570
#else
571 9d85cbd9 Fabrice Bellard
    return (int)(rint(x));
572 ea937d01 Fabrice Bellard
#endif
573 9d85cbd9 Fabrice Bellard
}
574
#endif
575
576 de6d9b64 Fabrice Bellard
#endif